Intruduction

Voters’ preference for political candidates from their own local area has been found consistently across time and space. Local candidates have been found to outperform their competitors in the United Kingdom (Arzheimer & Evans, 2012; Campbell & Cowley, 2014; Campbell et al., 2019; Evans et al., 2017), the United States (Lewis-Beck & Rice, 1983; Key, 1949; Tatalovich, 1975), Germany (Jankowski, 2016), Japan (Horiuchi et al., 2020), Ireland (Parker, 1982; Górecki & Marsh, 2014), New Zealand (Johnston, 1973), Estonia (Tavits, 2010), Poland (Górecki et al., 2022), and Norway (Fiva & Smith, 2017). - But why is this the case?

One group of explanations suggests that voters prefer local candidates because they believe that such candidates are more likely to act in the substantive interests of their local community (see, e.g., Shugart et al., 2005). This follows the assumptions of most normative models of democracy, according to which democratic accountability is ensured by voters who base their votes on political candidates’ policy positions or performance in office (see, e.g., Key, 1966; Schumpeter, 1943; Dahl, 1998). Voters infer that local candidates are better constituency servants and will be more receptive to the opinions of people like themselves (Campbell et al., 2019). Voting for a local candidate can therefore be seen as voters’ rational attempt to choose the candidate who best represents their own interests and those of their local area.

This explanation runs counter to Key’s (1949) interpretation. When he originally showed that voters favored local candidates and called it “friends-and-neighbors voting,” he saw it as a sign of “immature politics” (Key, 1949, 110). To him, the vote for the “home-town boy” was caused by factors that are unrelated to politics, and thus undermined electoral accountability. While the electoral advantage of local candidates is not literally caused by friends and neighbors voting for them, what Key alluded to can be thought of in contemporary terms as in-group favoritism. By parsing political candidates into categories such as “us” and “them,” voters guide their understanding of which candidates to vote for (Achen & Bartels, 2016). One such category may be their attachment to the local area, and voting for a local candidate may be a way of choosing “one of us” (Schulte-Cloos & Bauer, 2021; Collignon & Sajuria, 2018). Thus, understanding why voters prefer local candidates has important implications for understanding voter competence.

The present study extends previous work (see, e.g., Campbell et al., 2019; Schulte-Cloos & Bauer 2021; Collignon & Sajuria, 2018) by incorporating these two prominent explanations in a pre-registeredFootnote 1 conjoint experiment among a representative sample of the Danish population. This design allows me to experimentally manipulate a variety of different aspects of a candidate’s local attachments independently. In the study, I take advantage of the fact that in experiments, respondents make inferences based on the limited amount of information available to them (Dafoe et al., 2018; Hainmueller et al., 2014). By withholding information about certain aspects of a candidate’s local attachment from some respondents, I test how voters use information about candidates’ residences to make inferences about other aspects of their local attachment. In addition, I test whether the electoral advantage of local candidates is driven by voters with a strong identification with their local area.

I find that information about both candidates’ efforts to promote the interests of the local area and their efforts to conform to local group norms increases voters’ likelihood of voting for the candidate. However, voters appear to rely on information about candidates’ place of residence to make inferences about candidates’ substantive representation of local interests, and not to rely on information about candidates’ adherence to local group norms. Additional tests also show that the strength of voters’ own identification with their local area has no systematic effect on their preference for candidates who live nearby. Thus, this study finds no evidence that voters’ preferences for candidates who live nearby are driven by in-group favoritism.

Why Vote for a Local Candidate?

Explanations for the electoral advantage of local candidates can be grouped into three categories (Górecki et al., 2022). The first category emphasizes that residents are more exposed to information about local candidates, which gives the candidates an electoral advantage. However, recent experimental studies have documented that the electoral advantage of local candidates persists when campaign activities and local contacts are held constant across candidates (Campbell & Cowley, 2014; Campbell et al., 2019; Horiuchi et al., 2020). These studies find that information about where candidates were born, where they grew up, and where they currently live affects voters’ preferences for the candidate. Following Campbell et al. (2019), I refer to such information as the candidate’s “descriptive localism”. Knowing that voters rely on cues about candidates’ descriptive localism, however, does not explain why voters do so. What inferences do voters make about candidates who live or grew up nearby that makes them so universally appealing?

Two well-established theoretical explanations have been proposed in the literature: (1) self-interest and (2) in-group favoritism (see, e.g., Campbell et al., 2019; Collignon & Sajuria, 2018; Schulte-Cloos & Bauer, 2021). According to the first explanation, voters rely on information about candidates’ descriptive localism to make inferences about their behavioral localism. Campbell et al. (2019) define behavioral localism as the extent to which a politician acts in line with the interests and wishes of the voters themselves and others in their locality (Campbell et al., 2019, p. 938). The second explanation is that voters infer that candidates who live nearby are members of the local in-group and that voters prefer them because of in-group favoritism. To capture such aspects of candidates’ local attachment, I introduce the concept of symbolic localism. I define symbolic localism as the extent to which politicians signal membership in the local community of their constituents. Voters’ inferences about either aspect may explain why voters prefer local candidates.

Information about a candidate’s descriptive localism can thus be thought of as an informational shortcut (Lupia, 1994; Sniderman et al., 1991). Based on personal experience, voters may infer that political candidates with certain demographic characteristics are more likely to act and spend their time in certain ways. Information about a candidate’s age, occupation, or partisanship can give voters an idea of how the candidates are likely to spend their time. Despite how inaccurate such stereotypes may be, they help voters evaluate political candidates and, by extension, decide whether or not to vote for them (Cutler, 2002; Popkin, 1991, p. 63). Similarly, voters may make inferences about candidates’ behavioral localism or symbolic localism based on information about their descriptive localism.

Voters may infer that candidates who live close by will spend more of their time in office serving their own and their local area’s substantive interests. They may assume that local candidates better understand their local area and are better informed about their needs (Shugart et al., 2005, p. 939). Since support for political parties is clustered across geographic areas (Hansen & Stubager, 2017; Rodden, 2010), they may also think that local candidates hold similar views. Candidates can signal their behavioral localism by living up to such assumptions, being attentive to local needs, providing constituent services, and otherwise acting as a delegate of the local area (Campbell et al., 2019, p. 939). All of these activities are time-consuming and inevitably take time away from other activities, such as national policymaking or intraparty activities. It is therefore costly for candidates to signal high levels of behavioral localism.

Following the second explanation, voters may instead use information about candidates’ descriptive localism to identify candidates who match their own identities (Achen & Bartels, 2016; Schulte-Cloos & Bauer, 2021; Collignon & Sajuria, 2018, p. 313). According to social identity theory, individuals are motivated by a desire to maintain a positive self-conception (Tajfel & Turner, 1979, p. 40). Since certain groups are associated with higher or lower social status, individuals will seek to improve the social status of the groups to which they perceive themselves to belong. From a social identity perspective, you would expect voters to prefer local candidates because the candidate is part of the local in-group. This can be thought of as in-group favoritism – a psychological tendency to favor groups to which one is a part of (Tajfel & Turner, 1979, p. 28). Schulte-Cloos and Bauer (2021) thus also show that voters prefer descriptively local candidates even though the candidates have no chance of being elected and provide substantial interest representation to the local area.

Signaling symbolic localism to local voters is time-consuming. Candidates must show up and make a considerable effort to get to know their local community and demonstrate their commitment to constituents. While some candidates only show up in the constituency during election campaigns, others are so rooted in their local community that they know their constituents by name. How candidates can send signals about their symbolic localism varies according to the particular norms and values of the local community, and candidates must therefore tailor the signals they send to their particular local constituency (Fenno, 1978; Parker, 2015; Jacobs & Munis, 2019). This also makes it more difficult for those coming from elsewhere to emulate such signals. Based on these two perspectives, I hypothesize that:

  1. H1:

    Voters prefer local candidates because of their behavioral localism.

  2. H2:

    Voters prefer local candidates because of their symbolic localism.

To distinguish between these two explanations of voters’ preferences for descriptively local candidates, I exploit people’s tendency to fill in the blanks in the absence of information. In experimental studies, this phenomenon is known as a violation of information equivalence with respect to background features (Dafoe et al., 2018), as “masking,” “aliasing,” or “composite treatments” (Hainmueller et al., 2014). Respondents update their beliefs about adjacent attributes when presented with information about one aspect of a hypothetical scenario. This obscures the results of experimental studies, since the results may not be due to the object under study, but to adjacent attributes. However, it also allows me to get at the mechanism underlying voters’ preference for descriptively local candidates.

Based on the two theoretical perspectives outlined, I expect that voters who are provided with information about a candidate’s descriptive localism will make inferences about the candidate’s behavioral or symbolic localism (or both). In contrast, voters who are also provided with information about the candidate’s behavioral or symbolic localism will not make these inferences because the information renders such inferences redundant. Thus, the candidate’s descriptive localism should have less of an effect on their vote. In contrast, the effect of information about the candidate’s descriptive localism should be unaffected if voters do not make inferences about these particular adjacent attributes.

This approach has been used in previous studies of explanations of discrimination against minority candidates (Portmann, 2021), and to test whether voters use candidates’ descriptive localism to make inferences about their behavioral localism (Campbell et al., 2019). Campbell et al. find that among voters in the United Kingdom, the effect of descriptive localism is reduced when voters are provided with information about candidates’ behavioral localism (Campbell et al., 2019, p. 944). This finding suggests that voters make inferences about candidates’ behavioral localism based on their descriptive localism.

In this study, I randomly assign both information about a candidate’s behavioral localism and symbolic localism. Including both attributes allows me to distinguish between the two explanations. If voters prefer local candidates because of in-group favoritism, they may also use information about a candidate’s behavioral localism to make inferences about the candidate’s symbolic localism. Thus, my third hypothesis is that:

  1. H3:

    The effect of either a candidate's descriptive localism, behavioral localism, or symbolic localism is weakened when voters receive information about any of the other aspects.

The preceding hypotheses do not distinguish between how different voters process information about political candidates’ descriptive localism, behavioral localism, or symbolic localism differently. The apparent universality of the electoral advantage of local candidates in previous studies may justify this. Everyone lives somewhere, which at least to some extent anchors their self-interest and self-conception in a specific geographic context.

However, people move from time to time, which changes both the object of their self-interest and their place-based social identity. While self-interest may be highly flexible to such changes, the same cannot be said for social identities. Social identities are highly stable and change slowly over time (Huddy, 2001, p. 131). When people move to a new area, they are unlikely to act as long-time residents from day one. According to social identity theory, a prerequisite for individuals to engage in group behavior is that they identify with the group (Tajfel & Turner, 1979, p. 41). If voters do not identify with their local area, they can hardly be said to engage in in-group favoritism on behalf of the local area. People belong to many different social groups that have varying degrees of importance for their self-conception (Monroe et al., 2000; Tajfel et al., 1971). While previous studies suggest that voters identify at least to some degree with their local area (Wong, 2010, p. 77), the local area may not be a particularly important social identity for a substantial portion of the population. Voters may instead rely on other affiliations and favor a candidate based on shared gender, race, or partisan identity.

This provides another way to test whether voters favor local candidates out of in-group favoritism. By exploiting differences in the strength of voters’ place-based social identities, I can determine whether voters with strong local identities are more likely to favor local candidates:

  1. H4:

    Cues about political candidates’ local attachment are more important to voters who have a strong attachment to their local area than to voters who have a weaker attachment to the local area.

Design and Data

To distinguish between explanations of voters’ preferences for descriptively local candidates, it is necessary to be able to test how voters respond to different aspects of candidates’ local attachments separately and in combination. When voters choose a political candidate, they weigh a variety of different attributes of the candidate, such as their political attitudes and demographic characteristics. The choice to vote for a particular candidate is thus the combined effect of a number of different determinants. Conjoint experiments are ideal for disentangling such causal mechanisms because they enable the independent variation of a range of causal components (Hainmueller et al., 2014, p. 28). By withholding information about a candidate’s behavioral localism and/or symbolic localism from some respondents, I can identify the potential importance of these mediators of descriptive localism (Acharya et al., 2018). At the same time, this design allows me to closely simulate choices that reflect what voters otherwise encounter in the real world. This design is therefore ideal for identifying why voters prefer local candidates.

I conducted a conjoint experiment in a survey distributed through YouGov’s panel in Denmark. The sample consisted of 1,021 respondents. The respondents were selected based on quotas of the ideal distribution in the Danish population of gender, age, region, and education level. The total sample consisted of 10,210 unique respondent-candidate dyads (1021 \(\times\) 2 profiles \(\times\) 5 conjoint tasks). I do not use all 10,210 dyads in the analysis as 242 respondents were excluded due to inattention and up to 1175 dyads were excluded due to “don’t know” responses.

In each conjoint task, I presented respondents with two political candidates (Candidate A and B). The tasks were constructed as short vignettes that resembled texts from politicians’ “about the candidate” pages on a website or campaign materials. The candidates were described in five to seven sentences. In the sentences, I randomly assigned seven attributes to each candidate. The number of sentences varied as I omitted information about the candidates’ behavioral localism and symbolic localism from a random subsample.

The three main attributes in the vignettes were information about the candidates’ descriptive localism, behavioral localism, and symbolic localism. Information about the candidates’ descriptive localism was varied in the profiles along two dimensions: where the candidate lives and where the candidate was raised. Across countries, a large proportion of parliamentarians either grew up or currently live in their constituencies (Pedersen et al., 2008). I, therefore, prioritized capturing both of these notions of descriptive localism in the candidate profiles.

The information about the candidates’ behavioral localism was inspired by previous work by Campbell et al. (2019). The cues describe how the candidate typically allocates his or her time during a work week between local issues and national policies (Campbell et al., 2019). A candidate signals a high level of behavioral localism by spending most of their time on local issues. This operationalization thus captures the inherent time trade-off for candidates, as prioritizing local issues takes time away from other tasks.

In parallel, I operationalize information about candidates’ symbolic localism as an increasing time commitment to the local community. Does the candidate only show up at election time, or does the candidate spend more time in the community, attending community events, participating in the social life of local associations, and getting to know people by name?

Creating information cues that prime each explanation without priming the other is challenging. While a candidate who spends most of their time in office on local issues clearly signals behavioral localism, it may also inadvertently signal loyalty to the local in-group. Similarly, participation in local community associations may be seen as a way for the candidate to promote the substantive interests of the local area. Such a spillover could pose a challenge to the internal validity of the study since the results of one of the two primes could be interpreted as a consequence of the other. However, I do not find that the availability of cues about the candidate’s symbolic localism moderates the effect of cues about the candidate’s behavioral localism, which should limit concerns over such spillover. Nevertheless, my interpretation of the results still relies on the assumption that such spillover is minimal, or at least that the cues are much stronger primes of their respective phenomena than they are at priming the contrasting phenomena.

The remaining four control attributes are not of theoretical interest in this study, but improve the resemblance of the candidates to real-world candidates and provide a benchmark for the size of the effect estimates. These four attributes pertain to the candidates’ gender, age, occupation, and partisanship. The exact wording of each attribute level can be found in Table 1, and the full vignettes can be found in Appendix G.

To further improve the external validity of the study, I base the distribution of each of the control attributes on the marginal distribution in the target population (de la Cuesta et al., 2021), namely candidates for the Danish Parliament (Folketinget) in 2019. I rely on Statistics Denmark’s records of gender, age, occupation, and partisanship of candidates for Folketinget for the marginal distribution of each of the control attributes (Statistics Denmark, 2019). The main attributes are uniformly distributed to ensure sufficient power in each category.

The dependent variable in the experiment is the respondents’ attitudes toward the candidates. I measured respondents’ attitudes with two items. The first item is the question “How likely is it that you would vote for Candidate A [B]?”, which is scored on a likert scale ranging from “very unlikely” (1) to “very likely” (5). I have rescaled this variable to range from 0 to 1, with high values indicating that respondents say they are very likely to vote for the candidate. The second outcome measure is a forced-choice question. In Appendix C, I repeated the analysis using the forced-choice question as an alternative outcome measure. The results in Appendix C closely mirror those reported here.

I pre-registered the study’s hypotheses and analysis plan before the data was collected. By pre-registering the hypotheses, I am able to distinguish between the study’s predictions and postdictions (Nosek et al., 2018). Thus, pre-registering the hypotheses protects me from hindsight bias by limiting my ability to adjust the theories to fit the findings. The advantage of pre-registering the analysis plan is that it limits my leverage to p-hack and thus improves credibility (Elman et al., 2018, p. 37). When I analyzed the data to test the pre-registered hypotheses, I made some deviations from the pre-registered analysis plan. Following current advice in the literature, I report and justify these deviations in the text to allow the reader to assess their significance (Humphreys et al., 2013, p. 13). In addition, a full analysis, following the analysis plan in the pre-registration, can be found in Appendix A. The results in Appendix A do not contradict the results in the main analysis but lack statistical power and analytical clarity.

A drawback of survey experiments like the one conducted here is that they lack the stakes that characterize real-world elections. Therefore, I suspected that a substantial number of those surveyed would be inattentive. This poses a threat to the internal validity of the design, as the result may be due to arbitrary responses. To alleviate this concern, I included two candidates in each task, as previous studies suggest that paired designs improve respondent engagement (Hainmueller et al., 2015, p. 2400). Furthermore, I presented only five conjoint tasks to each respondent, although previous findings suggest that the number of tasks has a limited impact on the overall results (Bansak et al., 2018). Finally, I excluded respondents with unreasonably low response times (less than 3 minFootnote 2).

How voters evaluate political candidates is shaped by the institutional structure of the surrounding society. Previous studies on the importance of a candidate’s place of residence have mainly been conducted in countries with single-member districts (Key, 1949; Tatalovich, 1975; Lewis-Beck & Rice, 1983; Gimpel et al., 2008; Campbell et al., 2019; however, see Górecki and Marsh 2014). In such electoral systems, candidates do not share blame or credit for local policies with other local candidates. They are therefore uniquely positioned to build a personal vote that casts them as the champion of the local area (Pedersen & VanHeerde-Hudson, 2019, p. 19). However, single-member districts usually do not allow voters to choose between different candidates of the same party. Thus, a candidate’s place of residence must outweigh partisanship if it is to change people’s votes. As a result, previous studies have often omitted information on candidate partisanship or focused on primaries. However, in real-world settings, a candidate’s partisanship is almost an integral part of the candidate’s name and is highly relevant to people’s vote choices. Suppressing this piece of information limits the external relevance of any result, as partisanship may suppress the importance of localism in real-world settings.

By conducting the study in Denmark, I can alleviate these concerns. The Danish electoral system for the Folketing is a two-tier PR system with multi-member districts (Elklit, 2020). While the upper-tier mandates are distributed on a national scale to ensure proportionality, the lower-tier mandates are allotted to each geographical region based on population. This ensures some form of geographical representation. While it is up to the local party branches to decide how to select candidates from the party list, most use some form of open list where voters can influence which candidate from the party is elected (Elklit, 2020, p. 69). Open lists give politicians an incentive to compete with other partisans in the same region to obtain as many of the party’s votes as possible and climb the party’s list (Shugart et al., 2005, p. 438). Candidates will therefore try to cast themselves as the local champion, similar to candidates in single-member districts. In addition, the multiparty system means that the distances between parties are less pronounced. The hurdle that voters have to overcome to vote against their partisanship for a local candidate is thus smaller, and doing so does not necessarily mean that they are voting for a party with a different candidate for the post of prime minister.

Finally, previous studies have indicated that voters prefer candidates who share their demographic characteristics or partisan leanings (Cutler, 2002). To improve the efficiency of the estimates, I, therefore, include control variables indicating whether the respondents are of a similar age as the candidate (± 5 years), share the candidate’s gender, have the same occupation as the candidate, identify with the same political party, or identify with a party that promotes the same candidate for the post of prime minister.

Table 1 Distribution of attributes of candidates in conjoint task
Full size table

Preferences for Local Candidates

To explain why voters prefer local candidates, I first test whether voters actually prefer candidates who grew up in their local area, live in their local area, spend their time in office on local issues, and devote time to the local community. To do this, I estimate two quantities of interest: the marginal means (MM) for each attribute in the candidate profiles and the average marginal component effect (AMCE) of each candidate attribute. The MM is a descriptive statistic that captures the average outcome score for all candidates with the given attribute level, ignoring all other attributes (Leeper et al., 2020). The AMCE, on the other hand, captures the marginal effect of a candidate with a given attribute level relative to a reference category, averaged over the joint distribution of the remaining attributes. Since the candidate attributes are randomly assigned, it can be interpreted as the causal effect of having a particular attribute level relative to a reference category (Hainmueller et al., 2014). In Fig. 1, I report the MM and AMCE of the candidate attributes in the conjoint experiment. All standard errors are clustered at the respondent level.

As can be seen in the first row of the panels of Fig. 1, I find that, on average, respondents rate a candidate who lives in their local area 2.2 percentage points (95% CI 0.8–3.6) higher than a candidate who lives elsewhere, on a scale from 0: very unlikely to vote for a candidate to 1: very likely to vote for the candidate. This is even more evident when I use the forced-choice dependent variable (see Appendix C). Voters prefer candidates who live in their local area even when I control for other aspects, such as the candidate’s behavioral localism, symbolic localism, and partisanship. Meanwhile, where the candidate grew up does not seem to have a substantial effect on the respondents’ ratings. It thus seems to be the candidate’s current affiliation with the local area that matters to voters. In what follows, I, therefore, focus on the former operationalization of descriptive localism.

In the second and third rows of panels in Fig. 1, I also find that information about the candidate’s behavioral localism and symbolic localism matters for the respondents’ ratings. Respondents rate candidates 4.2 percentage points (CI 2.2–6.2) higher, on average, when they are told that the candidate spends most of their time in office on local issues than when they are not told about the candidate’s behavioral localism. I also find that the less time the candidate spends working on local issues, the less likely respondents are to say they would vote for the candidate.

Furthermore, I also find that voters prefer candidates who signal their symbolic localism by being involved in the local community. Telling respondents that the candidate knows people’s names and is active in local associations increases respondents’ ratings of the candidate by 4.1 percentage points (CI 2.0–6.1) compared to not disclosing information about the candidate’s symbolic localism. In contrast, respondents’ ratings of the candidate are 2.5 percentage points (CI 4.7–0.3) lower when they are told that the candidate is not regularly seen in the local area.

The substantial sizes of the effects of the candidate’s descriptive localism, behavioral localism, and symbolic localism are similar in size and non-trivial. The effects are larger than the effect of the other demographic cues that might otherwise be hypothesized to have an effect, such as the candidate’s gender, occupation, and age. While differences in outcome measures and treatment wordings prohibit a direct comparison to previous studies from the United Kingdom and Japan, the effects seem to be of a similar magnitude (Campbell et al., 2019; Horiuchi et al., 2020). However, the effects are dwarfed by the effect of identifying with the same party as the candidate (25.5 percentage points, CI 22.9–28.0) or with a party that supports the same candidate for prime minister (17.0 percentage points, CI 14.8\(-\)19.1).

These results show that the respondents do prefer local candidates. They are more likely to say that they would vote for a candidate if the candidate lives in their local area (descriptive localism), spends most of their time in office on local issues (behavioral localism), and is engaged in the local community and active in local associations (symbolic localism).

Fig. 1
figure 1

Estimated MM and AMCE of each candidate attribute level. All AMCEs are estimated compared to the baseline level (ref) of the attribute. Standard errors clustered at the respondent level. Bars show 95% confidence intervals. In the left panel, the bold vertical grey line indicates the average score on the outcome variable. Models include controls for correspondence between attributes of respondent and candidate characteristics with regard to gender, age, occupation, and partisanship

Full size image

Interactions Between Cues About Candidates’ Local Attachment

Voters often do not know how candidates distribute their time between local and national issues and the extent to which they invest time in the social life of the local community. According to H3, voters instead use information about a candidate’s place of residence to make inferences about these adjacent attributes. To test this, I compare the MM of descriptive localism in the subgroup that is informed about the candidates’ behavioral localism (or symbolic localism) to its MM in the subgroup that is not informed about the candidates’ behavioral localism (or symbolic localism).Footnote 3 I expect that respondents who are unaware of the candidate’s behavioral localism (or symbolic localism) will rely more heavily on information about the candidate’s descriptive localism.

In Fig. 2, I test H3 by separately estimating the interaction of each pair of cues about the local attachment of the candidate. Panel A shows how the availability of information about candidates’ behavioral localism affects the importance of information about the candidates’ descriptive localism. Consistent with H3, I find that voters rely more on information about the candidate’s place of residence when they do not know how the candidate will divide their time in office between local and national issues. In the subgroup that is not told about the candidate’s behavioral localism, respondents rate the candidates who live in their local area 4.9 percentage points (CI 1.8–7.9) higher than a candidate who lives elsewhere. Meanwhile, the same difference in effect is reduced to 1.5% points (CI 3.4 smaller–0.1 larger, p = 0.06) when respondents know about the candidate’s behavioral localism—a difference in the effect of 3.2 percentage points (CI 6.8 smaller–0.3 larger, p = 0.08). While the difference in effects does not reach conventional levels of statistical significance when I exclude inattentive respondents, the results still suggest that voters make inferences about candidates’ behavioral localism based on their descriptive localism. When faced with a candidate who resides elsewhere, voters do not make favorable assumptions about his or her behavioral localism. In contrast, they make favorable assumptions about the behavioral localism of candidates who live nearby. This result supports H3 as the effect of candidates’ descriptive localism is attenuated when voters receive information about candidates’ behavioral localism.

Fig. 2
figure 2

Interaction between different levels of descriptive localism, behavioral localism, and symbolic localism. Black points indicate that information about the candidate’s behavioral/symbolic localism was available to the respondent. Panel A = descriptive localism \(\times\) behavioral localism. Panel B = descriptive localism \(\times\) symbolic localism. Panel C = behavioral localism \(\times\) symbolic localism. Standard errors are clustered at the respondent level. Bars show 95% confidence intervals. P-values indicate the statistical significance of the difference in the effect of the attribute on the respondent’s likelihood of voting for the candidate relative to the reference category for that attribute

Full size image

Panel B shows how the effect of information about the candidate’s residence is affected by the availability of information about the symbolic localism of the candidate. In contrast to the preceding analysis, the effect of the candidate’s place of residence on the respondents’ likelihood to vote for the candidate does not seem to be affected by the availability of information about the candidate’s time commitment to the local in-group. While the respondents on average prefer candidates when they are informed about the symbolic localism of the candidate, this effect is not statistically significantly different for locally residing candidates and for those that reside elsewhere (0.2 percentage points difference CI 3.3 lower\(-\)3.8 higher). This result does not support H3 with respect to the symbolic localism of the candidate, as the candidate’s descriptive localism retains its explanatory power in the presence of information about the candidate’s inclination to engage in the local community. However, due to the limited statistical power and as indicated by the confidence interval, this study cannot rule out the possibility that there is a non-negligible difference in the effect of symbolic localism between candidates who live in the local area and those who live elsewhere. Thus, clear evidence of the absence of an effect of interest would require either a substantially larger sample or a less externally valid design that leaves more variation to be explained by descriptive localism and symbolic localism.

Finally, in Panel C of Fig. 2, I test whether respondents make inferences about how candidates divide their time in office between local and national issues based on information about the candidate’s symbolic localism. Here, I find no evidence that the availability of information about the candidate’s tendency to be involved in the local community affects the effect of information about the candidate’s behavioral localism. These results further suggest that H3 does not appear to hold with respect to this study’s operationalization of a candidate’s symbolic localism. However, it should also limit our concern that there is a masking between the operationalizations of behavioral localism and symbolic localism in this study.

While voters prefer candidates who spend a lot of time in the local community, this does not seem to be the reason why they prefer candidates who live in their local area. Instead, a candidate’s symbolic localism seems to be a separate dimension on which voters evaluate political candidates.

Place-based Social Identity and Preference for Local Candidates

To further test whether voters’ preference for candidates who live in their local area is caused by in-group favoritism, I estimate whether the strength of voters’ place-based social identity conditions their preference for local candidates (H4). I measured the strength of respondents’ identification with their local area prior to the conjoint tasks. Highly stable entities such as place-based social identities are difficult to manipulate experimentally. Previous studies have attempted to use priming exercises to increase the salience of local identities (see, e.g., Hopkins, 2018, p. 191). As described in Appendix E, in this study a random subsample was exposed to such a priming exercise. The exercise increased respondents’ scores on the local identity measure by 3.1 percentage points (CI \(-\)0.0\(-\)6.1). While the priming exercise increased the strength of the respondents’ place-based social identities, the substantial effect is too small to allow me to test H4 experimentally. Therefore, consistent with the pre-registration, I have relied on the observational measure of local attachment to test H4.

I measure the strength of respondents’ identification with their local area with three items that are based on items previously used in studies of place-based social identities (Huddy & Khatib, 2007). Since I cannot identify the respondents’ local area, I have slightly modified the wording. In doing so, I also avoid making assumptions about respondents’ conceptions of what constitutes their subjective local area (Wong et al., 2012). The item wordings are as follows (see Appendix E or G for the unaltered and Danish translations):

  1. 1)

    To what extent do you see yourself as typical of people from your local area?

  2. 2)

    How often do you say “we” instead of “they” when you talk about events, people, and other conditions in your local area?

  3. 3)

    To what extent do you feel attached to your local area?

From these three items, I create an additive index of respondents’ identification with the local area, which I re-scale to range from 0 to 1. The index is reliable (Cronbach’s alpha = 0.74), and the three items appear to capture the same underlying phenomenon (correlations of 0.42, 0.55, 0.53).

I define respondents as having a strong identification with their local area if they score 0.6 or higher on the index (N=234) and as having a weak identification with their local area if they score less than 0.2 (N=67). The remaining respondents (N=468) are coded as having a medium level of identification with their local area. These thresholds capture meaningful levels in the index, with high scores including responses such as “to a great extent” and “often,” while the low scores never include these responses. In Fig. 3, I report the MM and the difference in MM for the different aspects of candidates’ local attachment for these three subgroups.

Fig. 3
figure 3

Estimated MM and difference in MMs for candidates’ descriptive, behavioral, and symbolic localism by respondents’ attachment to their local area. Black points = strong local attachment, grey points = medium local attachment, and white points = weak local attachment. Medium local attachment is the reference category in the right panel. Standard errors are clustered at the respondent level. Bars show 95% confidence intervals

Full size image

At most candidate attribute levels, there are only small differences between the evaluations of candidates by respondents with moderate and strong local identification, while respondents with weak local identification generally appear to be more skeptical of all candidates. Consistent with H4, I find that respondents with strong local identification rate candidates who spend their time in office primarily on local issues 10.8 percentage points (CI 3.8\(-\)17.8) higher than respondents with weak local identification. However, in contrast to H4, respondents with medium or strong local identification are also more favorable toward candidates who are rarely seen in the local area. Thus, the skepticism of respondents with a weak local identification seems to be independent of the candidate’s attributes.

In Appendix C, I replicate the analysis using the forced-choice question as the dependent variable. By design, the forced-choice measure has a mean of 0.5 within each respondent, which eliminates any differences between respondents at baseline. In this analysis, I find no difference in the three groups’ preferences for local candidates across the different candidate attributes.

Overall, these results do not support the hypothesis that voters’ preference for local candidates is a matter of in-group favoritism. While weak identification with the local area is associated with lower support for candidates at baseline, this difference seems to be due to differences in the composition of the groups I am comparing.

Conclusion

Political candidates who live nearby can possess many different attributes that appeal to voters. Using a conjoint experiment, I test whether voters prefer descriptively local candidates because of behavioral localism and symbolic localism. I find that while voters prefer candidates who live nearby, the effect of knowing where candidates live is much less pronounced when voters are informed about candidates’ behavioral localism. Thus, voters appear to prefer candidates who live nearby in part because of their expectations about how the candidate will spend his or her time in office.

I also find that voters prefer candidates who spend time being visible in the social life of the local community. However, I do not find that such preferences are related to voters’ preference for candidates who reside in their local area, even though these two phenomena may be empirically correlated. In addition, I do not find that voters’ identification with their local area conditions the electoral advantage of descriptively local candidates. The results of this study thus suggest that voters’ preference for candidates who spend considerable time signaling their membership in the local community appears to operate on a separate dimension from voters’ preference for candidates who live nearby.

In this study, I do not exhaust the explanatory power of a candidate’s descriptive localism. Other operationalizations of symbolic localism, or entirely different explanations than those explored here may help explain why voters prefer candidates who live nearby. Moreover, local candidates are likely to enjoy an additional informational advantage in the real world. Here, they can take advantage of their easy access to the local electorate and increase their exposure through daily interactions. This may also explain why Schulte-Cloos and Bauer (2021) find that voters favor candidates even when they have no chance of winning and representing the interests of the local area.

While the institutional structure in Denmark conditions the results, the mechanism studied behind voters’ preference for local candidates is likely to be generalizable. Campbell et al. (2019) find similar results on the importance of behavioral localism in the United Kingdom, suggesting that the findings are not an artifact of the electoral or party system in either Denmark or the United Kingdom.

These findings are encouraging for the prospects of democratic accountability. Most democratic ideals assume that voters are responsive to the behavior of elected officials (Key, 1966, p. 61). Key deplored friends-and-neighbors voting precisely because it suggested to him that voters blindly follow irrelevant group loyalties (Key, 1949, p. 110). The results of this study, however, suggest that voters’ preferences for local candidates are based on more rational motivations. Voters seem to prefer local candidates because they seek a candidate who will act in accordance with the substantive interests of their local area. This seems to be an often justified assumption, as previous studies have shown that local politicians do promote the interests of their own local area once elected (Tavits, 2010; Fiva & Halse, 2016; Binderkrantz et al., 2020; Carozzi & Repetto, 2016).